本文作者开创了多目标协同分割算法在现有的数据集中分割效果良好，用途很广泛。文的优点在于：由于标号增长的准确性在图形转换半监督学习框架中主要依赖于图形节点的相似性有多可靠，在Gunhee Kim ，Eric P. Xing[9]等人提出的多前景协同分割中节点代表着图像区域，它们的相似性既不是很容易区别又不是特别的稳定，尤其是不同图片中相同物体很大的外观差别，分割成属于不同物体的相似性有可能大于分割成相同的物体。这样本文提出结合全局连通约束图形转换半监督学习框架。每个约束项能够看成是一个线性不等式，基于此，我们开创了切平面算法，它能够解决线性不等式约束项和标号增长的凸一元二次方程之间的迭代，寻找最大违反约束项。
SHOW FULL COLUMNS FROM `jrk_downrecords` [ RunTime:0.000922s ]
SELECT `a`.`aid`,`a`.`title`,`a`.`create_time`,`m`.`username` FROM `jrk_downrecords` `a` INNER JOIN `jrk_member` `m` ON `a`.`uid`=`m`.`id` WHERE `a`.`status` = 1 GROUP BY `a`.`aid` ORDER BY `a`.`create_time` DESC LIMIT 10 [ RunTime:0.102370s ]
SHOW FULL COLUMNS FROM `jrk_tagrecords` [ RunTime:0.013274s ]
SELECT * FROM `jrk_tagrecords` WHERE `status` = 1 ORDER BY `num` DESC LIMIT 20 [ RunTime:0.009113s ]
SHOW FULL COLUMNS FROM `jrk_member` [ RunTime:0.006283s ]
SELECT `id`,`username`,`userhead`,`usertime` FROM `jrk_member` WHERE `status` = 1 ORDER BY `usertime` DESC LIMIT 10 [ RunTime:0.003128s ]
SHOW FULL COLUMNS FROM `jrk_searchrecords` [ RunTime:0.001723s ]
SELECT * FROM `jrk_searchrecords` WHERE `status` = 1 ORDER BY `num` DESC LIMIT 5 [ RunTime:0.003631s ]
SELECT aid,title,count(aid) as c FROM `jrk_downrecords` GROUP BY `aid` ORDER BY `c` DESC LIMIT 10 [ RunTime:0.014257s ]
SHOW FULL COLUMNS FROM `jrk_articles` [ RunTime:0.001123s ]
UPDATE `jrk_articles` SET `hits` = 2 WHERE `id` = 324361 [ RunTime:0.022861s ]
